Subtask Assignment for an Artificial Intelligence Task

1. A method, including: receiving an artificial intelligence (AI) computation task to be performed; identifying a set of client devices that are available in a local environment to participate in performing the AI computation task; identifying attributes of individual client devices of the set of client devices, including identifying security specifications for the set of client devices; dividing the AI computation task into individual subtasks; identifying hardware specifications for the individual subtasks; assigning, based on the attributes, the individual subtasks of the AI computation task to different client devices of the set of client devices in the local environment, including: identifying, from multiple different defined security levels, a security level for a first subtask of the individual subtasks; identifying a computation resource for performing the first subtask of the individual subtasks; and assigning the first subtask to a first client device of the set of client devices based at least in part on identifying that the first client device has a security specification corresponding to the security level for the first subtask, the computation resource for performing the first subtask, and a hardware specification for the first subtask; and performing the AI computation task by executing the individual subtasks of the AI computation task at the different client devices.

2. The method of claim 1, wherein said identifying the set of client devices comprises inspecting a system map that identifies client devices that are available in the local environment for participating in the AI computation task.

3. The method of claim 1, wherein said identifying attributes of the individual client devices comprises identifying, for each of the individual client devices, hardware resources or security attributes.

4. The method of claim 1, wherein at least one of the individual subtasks includes a subtask criteria for executing the individual subtasks of the AI computation task, and wherein the method further comprises assigning the at least one subtask to a particular client device based on identifying that attributes of the particular client device correspond to the subtask criteria.

5. The method of claim 4, wherein the subtask criteria specifies one or more of a computing resource criterion for the at least one subtask.

6. The method of claim 1, wherein at least one of the individual subtasks represents one or more layers of an AI model.

7. The method of claim 6, wherein said assigning comprises assigning execution of the one or more layers of the AI model to a particular client device.

8. The method of claim 1, wherein the individual subtasks each represent different layers of an AI model, and wherein said assigning comprises assigning a different layer of the AI model to each of the different client devices.

9. The method of claim 1, wherein the assigning the individual subtasks of the AI computation task to the different client devices of the set of client devices comprises: assigning a first subset of the individual subtasks for execution by a first set of the individual client devices; assigning a second subset of the individual subtasks for execution by a second set of the individual client devices; and merging results of the first subset of the individual subtasks and the second subset of the individual subtasks into a merged result of the AI computation task.

10. The method of claim 1, further comprising: identifying, based at least in part on a change in an attribute of the first client device, that the first subtask is to be reassigned; and reassigning the first subtask to a different client device of the set of client devices to cause the first subtask to be executed by the different client device.

11. A system including: one or more processors; and one or more computer-readable storage memory storing instructions that are executable by the one or more processors to: receive an artificial intelligence (AI) computation task to be performed; identify a set of client devices that are available in a local environment to participate in performing the AI computation task; divide the AI computation task into individual subtasks; identify hardware specifications for the individual subtasks; assign the individual subtasks to different client devices of the set of client devices in the local environment, including to: identify, from multiple different defined security levels, a security level for a first subtask of the individual subtasks; identify a computation resource for performing the first subtask of the individual subtasks; identify security specifications for the different client devices; and assign the first subtask to a first client device of the different client devices based at least in part on identifying that the first client device has a security specification for the first client device corresponding to the security level for the first subtask, the computation resource for performing the first subtask, and a hardware specification for the first subtask; and perform the AI computation task by executing the individual subtasks of the AI computation task at the different client devices.

12. The system of claim 11, wherein to identify the set of client devices that are available in the local environment to participate in performing the AI computation task includes to inspect a system map that identifies devices in the local environment that include attributes indicating that the devices are capable of executing one or more of the individual subtasks.

13. The system of claim 11, wherein to assign individual subtasks of the AI computation task to the different client devices includes to: determine device capabilities of the set of client devices; determine subtask criteria for the individual subtasks; and assign the individual subtasks to respective client devices based on client devices that exhibit capabilities that correspond to the subtask criteria.

14. The system of claim 13, wherein to assign the individual subtasks of the AI computation task to the different client devices includes to: assign the individual subtasks to a second set of the client devices based on identifying that the second set of the client devices is likely to execute the individual subtasks faster than a first set of the client devices.

15. The system of claim 14, wherein to assign individual subtasks of the AI computation task to the different client devices includes to: determine first subtask output from the first set of the client devices and second subtask output from the second set of the client devices; and merge the first subtask output and the second subtask output into a task result for the AI computation task.

16. The system of claim 11, wherein to determine that the AI computation task is to be performed includes to determine that the AI computation task is to be performed using an AI model with multiple layers, and wherein to assign individual subtasks of the AI computation task to the different client devices of the set of client devices includes to assign execution of different respective layers of the AI model to different respective client devices.

17. The system of claim 11, wherein the instructions are executable by the one or more processors to determine that a particular subtask includes capturing sensor data via one or more sensor, and wherein to assign the individual subtasks of the AI computation task to different client devices includes to assign the particular subtask to a client device identified as having a capability of capturing the sensor data.

18. A method, including: receive an artificial intelligence (AI) computation task to be performed via an AI model with multiple layers; identifying a set of client devices that are available in a local environment to participate in performing the AI computation task; dividing the AI model into different individual layers; identifying hardware specifications for the different individual layers; assigning the different individual layers of the AI model to different client devices of the set of client devices in the local environment, wherein a particular layer of the AI model includes an input of sensor data, and wherein said assigning the different individual layers of the AI model to the different client devices comprises: identifying sensor capabilities of at least some client devices of the set of client devices; identifying a computation resource for performing the particular layer of the different individual layers; and assigning the particular layer to a client device that is identified as being capable of capturing the sensor data and based on a hardware specification for the particular layer and the computation resource for performing the particular layer of the different individual layers; and performing the AI computation task by executing the different individual layers of the AI computation task at the different client devices.

19. The method of claim 18, wherein the AI model comprises an artificial neural network, and wherein the multiple layers comprise an encoding layer, one or more intermediate layers, and a decoding layer.

20. The method of claim 19, wherein said assigning the different individual layers of the AI model to the different client devices comprises assigning the encoding layer to a first client device, the one or more intermediate layers to a second client device, and the decoding layer to a third client device.